Known quantities
HP has substantial experience in multi-core technology, with its own balanced dual- and quad-core systems that take full advantage of the aggregate computing power now available. The HP Message Passing Interface Library is optimized for both shared memory and cluster systems on HP ProLiant and Integrity servers, and multi-core application scheduling is accommodated within HP’s XC Linux® cluster software. HP also offers Unified Parallel C (UPC), a development environment that has been available for several years and offers extensions to the C compiler.
It’s a strong legacy, but HP recognized the need for resources to help the industry address the challenges of multi-core processing. That’s why it has announced the HP Multi-core Optimization Program, part of the HP “Catalysts for High Performance Computing [HPC] Innovation” initiative, which drives adoption of emerging HPC technologies through advanced development projects and customer collaborations. The Multi-core Optimization Program addresses the major challenges that face application developers in reworking applications, or writing new ones, to run on multi-core processing systems.
The HP Multi-core Optimization Program aggregates HP’s resources with those of a range of technology partners including research institutions, strategic HP customers, hardware partners and development tool providers. These partners together with HP provide tools, technologies and best practices that application developers can use in optimizing for multi-core processors, including compiler suites, multi-core performance analytics, business intelligence software, multi-core development platforms, mathematical algorithm libraries and others. Nearly two dozen partners already contribute, with more coming on-stream.
Tools for the job
The Toolkit sorts multi-core issues into categories such as application development and compilers, performance and tuning tools, job scheduling and libraries, and more. The Toolkit Web site is subdivided according to these categories and within each, users will find links to HP, partner and open source products and tools. There is also a section on best practices for each category, offering white papers, efficiency processes, demos and multi-core ‘do’s and don’ts.’ Topics include power consumption, application performance on HP servers that use multi-core processors, measuring power against performance as cores are added, and more.
HP tests multi-core performance with each of its partners to confirm that products perform in multi-core systems, and also does interoperability testing so customers know that products and systems will be compatible. In support of the testing program HP has authored a Multi-core Toolkit Taxonomy white paper, to help partners optimize system testing.
Future plans include a reference model for the development environment, research into non-traditional uses for multi-core systems, and dedicated cores designed for better security, ease of management and other mission-critical roles.
“Enterprises may not see the multi-core problem just yet, but they will when they start going beyond quad-core,” Korf says. “The real issue is how to get ready. HP’s approach is to address the issue holistically, with a single source of information and resources to cope with the challenges of optimizing applications for multi-core processors.”
|
Printable version
“Moore’s law has traditionally allowed processor performance to double about every two years through increases in clock speed, without dramatically increasing power consumption,” says Dave Korf, HP Multi-core Marketing Program Manager. “But now, even small increases in clock speed come with large increases in power consumption. Adding cores has become the favored approach to increasing performance: it boosts aggregate compute power without demanding a huge jump in power consumption.”